Fix sqlite decoder inference

[Karakatiza666]

The SQLite examples stop type-checking on any nightly from nightly-2026-08-30-34e7489 onward.

── ✗ missing method ─ examples/sqlite-basic.roc:65:8
This is trying to dispatch a method named find_first_index on an unresolved
type variable, but unresolved type variables have no methods.
    id = Sqlite.i64("id")(cols)(stmt)?

Same error at examples/sqlite-everything.roc:204 and :219.

Root cause

lookup_value! found a column with cols.find_first_index(...). The decoder combinators are deliberately unannotated, so the compiler has not worked out what cols is at that point, and it carries the question into whatever decoder an application writes, where it still cannot be answered. roc-lang/roc#10984 stopped those questions sharing an answer between calls, so a decoder with a name of its own asks about cols once per column and gets nowhere.

Only named decoders break. The inline rows: |cols| |stmt| ... form waits for query_many! to say what cols is, which is why the count is three and not more.

Applications cannot work around it. query_many! unwraps the opaque Stmt and hands the decoder a Host.SqliteStmt, and Host is not in the platform's exposes list, so there is no annotation an app is able to write.

Fix

• call the builtin by name in lookup_value!: List.find_first_index(cols, ...). That says which type the lookup comes from, so cols settles as a List(Str) where it is asked about and nothing is left for the application to answer. platform/Path.roc:558 already reads the same lookup this way
• add an expect covering a row decoder with a name of its own, which is the shape that breaks
• move CI from nightly-2026-08-23-fb208ba to nightly-2026-08-30-34e7489. #10984 does not exist at the old pin, so neither the bug nor the test guarding it shows up there. I will drop bumping the nightly if you would rather move the pin separately.

No signature changes, so src/roc_platform_abi.rs and ci/regenerate_glue.sh are untouched: lookup_value! is module-private, sits outside the Sqlite :: [].{} block, and appears in no requires, provides or hosted block.

Verification

The test was checked against the bug on nightly-2026-08-30-34e7489:

platform/Sqlite.roc	roc test platform/main.roc
this PR	exit 0
fix reverted, test kept	exit 1

Note that roc test prints All (208) tests passed even when the type error fires, so the exit code is the signal here, not the summary line.

./scripts/test.py was not run. The build and run stages, and the two test_spec.json cases per SQLite example, are for CI to confirm. The change is Roc-only and moves no program output.

roc fmt --check already fails on platform/Cmd.roc, platform/InternalSqlite.roc and platform/Url.roc on both nightlies, unrelated to this change and untouched by it.